R@H Scientists/Coders: An analysis of the Rosetta binaries...

Message boards : Number crunching : R@H Scientists/Coders: An analysis of the Rosetta binaries...

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · Next

AuthorMessage
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1909
Credit: 8,818,406
RAC: 11,172
Message 78483 - Posted: 20 Jul 2015, 6:47:43 UTC - in response to Message 78478.  

or accept SSE3 as the baseline


+1
From now (without optimization) to SSE4.2, i think SSE3 is an acceptable compromise (not to loose too many cpu).

ID: 78483 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1909
Credit: 8,818,406
RAC: 11,172
Message 78484 - Posted: 20 Jul 2015, 7:19:34 UTC - in response to Message 78482.  

Rosetta must already parses this file to determine PLATFORM (Windows or Linux) and what kind of jobs I want to accept.
If they can handle TWO platforms, then this platform dispatcher could be expanded to include additional granularity of FEATURES supported.


The Boinc scheduler can "route" the work properly according to the cpu capabilities.
ID: 78484 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1829
Credit: 118,331,123
RAC: 25,769
Message 78486 - Posted: 20 Jul 2015, 9:02:06 UTC - in response to Message 78484.  
Last modified: 20 Jul 2015, 9:02:18 UTC


The Boinc scheduler can "route" the work properly according to the cpu capabilities.


My first thought was "is that going to require a server software update?" but I can't see any mention of "CPU" or "SSE" in the server updates list: http://boinc.berkeley.edu/trac/wiki/ServerUpdates

Server software upgrade info here: http://boinc.berkeley.edu/trac/wiki/ToolUpgrade
ID: 78486 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
rjs5

Send message
Joined: 22 Nov 10
Posts: 273
Credit: 22,294,386
RAC: 10,897
Message 78487 - Posted: 20 Jul 2015, 10:26:12 UTC - in response to Message 78486.  
Last modified: 20 Jul 2015, 10:37:44 UTC


The Boinc scheduler can "route" the work properly according to the cpu capabilities.


My first thought was "is that going to require a server software update?" but I can't see any mention of "CPU" or "SSE" in the server updates list: http://boinc.berkeley.edu/trac/wiki/ServerUpdates

Server software upgrade info here: http://boinc.berkeley.edu/trac/wiki/ToolUpgrade



I don't think it requires any Server update. I think the feature already exists as part of BOINC and Rosetta is already using it.

RATHER than generating FAT BINARIES which determine the CPU capabilities at runtime, Rosetta already KNOWS what the requesting system "p_features" are and would simply direct the right jobs to CPUs with that capability.

20 lines of XML programming to match David's SSE4.2 binary could allow it go to SSE4.2 machines.

http://boinc.berkeley.edu/trac/wiki/BoincPlatforms

http://boinc.berkeley.edu/trac/wiki/AppPlan
ID: 78487 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1829
Credit: 118,331,123
RAC: 25,769
Message 78488 - Posted: 20 Jul 2015, 11:07:17 UTC - in response to Message 78487.  

ok, so the process is:

1. compile versions for: non-SSE, SSE2, SSE3, SSE4.1, SSE4.2 (some can probably be omitted depending on performance effects)

2. create the "plan_class_spec.xml" file, which will need the "cpu_feature" tag completing for each version. This page shows how to nest the versions: http://boinc.berkeley.edu/trac/wiki/AppVersionNew
It's not clear whether you have to use the "projected_flops_scale" tag to make SSE4.2 the preferred download, or whether you just put that first in the "plan_class_spec.xml" and it will work through to find the first version that suits the CPU.

3. Watch Rosetta throughput increase as if 10,000 computers were added (after testing on Ralph first obviously)!
ID: 78488 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Dr. Merkwürdigliebe
Avatar

Send message
Joined: 5 Dec 10
Posts: 81
Credit: 2,657,273
RAC: 0
Message 78489 - Posted: 20 Jul 2015, 19:45:55 UTC - in response to Message 78487.  



RATHER than generating FAT BINARIES which determine the CPU capabilities at runtime, Rosetta already KNOWS what the requesting system "p_features" are and would simply direct the right jobs to CPUs with that capability.


Yeah, definitely the better solution. The rosetta binaries are already fat enough because of the static linking.

Again, nice to see some progress :-)
ID: 78489 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1909
Credit: 8,818,406
RAC: 11,172
Message 78490 - Posted: 21 Jul 2015, 9:04:01 UTC - in response to Message 78479.  

VS2015 says: Build for iOS, Android, Windows devices, Windows Server or Linux


Here the list of new features (some of) and here new features of C++

ID: 78490 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1829
Credit: 118,331,123
RAC: 25,769
Message 78493 - Posted: 23 Jul 2015, 17:19:00 UTC

Has there been any movement on getting the Rosetta source code to rjs5 (presumably along with some test jobs to run)?

If getting the source license issues resolved in the short term, would it be easier to set up a linux virtualbox at the bakerlab with the necessary source code and test jobs that rjs5 can log into using teamviewer or similar?
ID: 78493 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
rjs5

Send message
Joined: 22 Nov 10
Posts: 273
Credit: 22,294,386
RAC: 10,897
Message 78494 - Posted: 24 Jul 2015, 22:18:26 UTC - in response to Message 78493.  

Has there been any movement on getting the Rosetta source code to rjs5 (presumably along with some test jobs to run)?

If getting the source license issues resolved in the short term, would it be easier to set up a linux virtualbox at the bakerlab with the necessary source code and test jobs that rjs5 can log into using teamviewer or similar?


I have not heard from anyone but that is not big surprise. I sent a note to David E K offering any help I could. I think that would mean that I would be operating under the supervision of David or whoever wants to take the responsibility. If I fill out the license, I would still expect to communicate with someone to return anything I find with prototyped changes.

{url]https://c4c.uwc4c.com/licenses/86[/url]


David is currently doing what I would be doing so far. The kind of constraints they would place on source code changes will determine the limits of any benefit.







ID: 78494 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1909
Credit: 8,818,406
RAC: 11,172
Message 78504 - Posted: 27 Jul 2015, 9:24:18 UTC - in response to Message 78493.  

Has there been any movement on getting the Rosetta source code to rjs5 (presumably along with some test jobs to run)?


July and August are holidays...
Waiting for September with SSE3 on Win 64 bit :-)

ID: 78504 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1909
Credit: 8,818,406
RAC: 11,172
Message 78524 - Posted: 31 Jul 2015, 18:06:44 UTC

You can also contact Sesef, a great programmer and a boinc expert
http://optos.sesef.pl/
ID: 78524 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1909
Credit: 8,818,406
RAC: 11,172
Message 78566 - Posted: 8 Aug 2015, 22:09:36 UTC - in response to Message 78524.  

You can also contact Sesef, a great programmer and a boinc expert
http://optos.sesef.pl/


And Peciack http://chomikuj.pl/rakowskipw/boinc

ID: 78566 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Chilean
Avatar

Send message
Joined: 16 Oct 05
Posts: 711
Credit: 26,694,507
RAC: 0
Message 78572 - Posted: 14 Aug 2015, 4:49:52 UTC

Any news on this? :)
ID: 78572 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1909
Credit: 8,818,406
RAC: 11,172
Message 78573 - Posted: 14 Aug 2015, 15:36:51 UTC - in response to Message 78572.  

Any news on this? :)


It's August. Sun, beaches, waves and mojitos!

P.S.
Sesef also creates yesterday an optimized version of Denis@Home
8x/10x speedup with SSE3!!

ID: 78573 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Chilean
Avatar

Send message
Joined: 16 Oct 05
Posts: 711
Credit: 26,694,507
RAC: 0
Message 78574 - Posted: 14 Aug 2015, 16:08:56 UTC - in response to Message 78573.  
Last modified: 14 Aug 2015, 16:14:07 UTC

Any news on this? :)


It's August. Sun, beaches, waves and mojitos!

P.S.
Sesef also creates yesterday an optimized version of Denis@Home
8x/10x speedup with SSE3!!


I just finished an WU under 10 minutes (7 minutes 57 seconds) using this optimized version.
Impressive (assuming the work done is legit).
ID: 78574 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1909
Credit: 8,818,406
RAC: 11,172
Message 78601 - Posted: 28 Aug 2015, 18:14:39 UTC

Some optimization news on Ralph
sse
3.61
ID: 78601 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1909
Credit: 8,818,406
RAC: 11,172
Message 78850 - Posted: 23 Sep 2015, 19:59:12 UTC

Any "optimization news"??
ID: 78850 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1909
Credit: 8,818,406
RAC: 11,172
Message 78866 - Posted: 28 Sep 2015, 10:10:18 UTC - in response to Message 78566.  

You can also contact Sesef, a great programmer and a boinc expert
http://optos.sesef.pl/


And Peciack http://chomikuj.pl/rakowskipw/boinc


And Crunch3r. He writes the new optimized app for linux in Denis@home's project (using AMD Libm)
ID: 78866 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Chilean
Avatar

Send message
Joined: 16 Oct 05
Posts: 711
Credit: 26,694,507
RAC: 0
Message 78872 - Posted: 30 Sep 2015, 14:17:52 UTC

Sesef just released another optimization app for DENIS. It's even faster now.
Crunch3r helped out with their Linux binary (which I'm also running).

Their WUs went from about 1 hour to under 6 minutes for completion.
ID: 78872 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Dr. Merkwürdigliebe
Avatar

Send message
Joined: 5 Dec 10
Posts: 81
Credit: 2,657,273
RAC: 0
Message 78873 - Posted: 30 Sep 2015, 16:58:29 UTC - in response to Message 78872.  

Sesef just released another optimization app for DENIS. It's even faster now.
Crunch3r helped out with their Linux binary (which I'm also running).

Their WUs went from about 1 hour to under 6 minutes for completion.

Hi, as you may have noticed we're only talking to ourselves here.
ID: 78873 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · Next

Message boards : Number crunching : R@H Scientists/Coders: An analysis of the Rosetta binaries...



©2024 University of Washington
https://www.bakerlab.org